Evaluating forest aboveground biomass estimation model using simulated ALS point cloud from an individual-based forest model and 3D radiative transfer model across continents

Abstract

Area-based approach (ABA) has been widely employed for estimating forest aboveground biomass (AGB) using airborne laser scanning (ALS) data. However, its scalability is limited due to challenges in model generalization across different forest types and regions. The selection of sensitive variables from ALS data is crucial for constructing robust forest AGB estimation models, yet this selection varies significantly among forest types and regions. Traditionally, assessing the influence of variable selection is hindered by the lack of accurate reference forest AGB values. Computer simulation-based method provides a perspective for exploring these challenges. This study employs an individual-based forest growth process model, FORMIND, coupled with a 3D radiative transfer model (RTM), LESS, to evaluate the transferability of ABA-based forest AGB estimation models and the generalization of ALS-derived variables. We used six virtual 3D forest scenes and two real-world forest sites, representing a range of global forest types, along with their simulated ALS data, to develop a forest AGB estimation model using the random forest algorithm, which allowed us to analyze the importance of various variables. We assessed model transferability through cross-comparison. Additionally, we validated the model using field plots and ALS data collected from two distinct regions. The results showed that the canopy surface area and volume extracted using the α-shape algorithm and parameters fitted from the Weibull distribution are vital variables when using ALS for forest AGB estimation across forest types and regions. Incorporating these variables into the model significantly improves the accuracy of forest AGB estimation. The optimized model achieved a R2 of 0.945, a RMSE of 34.22 t/ha, and a MAE of 20.53 t/ha. Our study not only deepens the understanding of the relationship between forest vertical structural metrics and AGB but also highlights the potential of computer simulation as a tool for refining the estimation of forest structural parameters.